Yeah I did something similar, mine were 2860MHz NB and 1980MHz NB @ 2420MHz CPU and 1980MHz HT. Many many tests, real tests. The video decoding/encoding proved it that CPU/Mem/HT clocks are the same. CPU-Z doesn't detect properly. Try and validate it with high NB, it'll fail like this: http://valid.x86-secret.com/show_oc.php?id=286546

All other software is reporting correctly, it is reading everything from the core registers themselves as it should. However, the core itself is telling its frequency as low values for some reason and causing accurate software and OSes to all error. Meanwhile, the actual frequencies remain unchanged.

About the NB speed. There is only one software where I've seen any perf. difference with high NB readings: CPU-Z latencies, which I don't know if its synthetic (based on CPU-Z readings) or real-time. I will ask Franck about it. Sandra does not show NB speed gain whatsoever, neither anything other including all popular benchmarks be it synthetic or real-time and I ran them all that I could, even WCG benchmark. Performance was exactly the same, even in encoding/decoding, compression//decompression and encrpytion/decryption. In fact, decoding encoding 13 videos which took 15 minutes 56 seconds, took exactly the same time for both, absolutely same.

Keep up your tests. Yes I'll probably have BE soon, haven't contacted my uncle to ask if he has any yet and can't until Monday when he comes back. Only Intel can change that, if their quads hit in early Jan, because then I'll want to play with them first instead.